LOOP Plugins

Local out of process (LOOP) plugins using github.com/hashicorp/go-plugin.

Packages

flowchart
    subgraph plugin-common/pkg
        loop
        internal[loop/internal]
        pb[loop/internal/pb]
        test[loop/internal/test]

        internal --> pb
        test --> internal
        loop --> internal
        loop --> test
    end
    
    grpc[google.golang.org/grpc]
    hashicorp[hashicorp/go-plugin]

    loop ---> hashicorp
    loop ---> grpc
    test ---> grpc
    internal ---> grpc
    pb ---> grpc
    hashicorp --> grpc

package loop

Public API and hashicorp/go-plugin integration.

package test

Testing utilities.

package internal

GRPC client & server implementations.

package pb

Protocol buffer definitions & generated code.

Communication

GRPC client/server pairs are used to communicated between the host and each plugin. Plugins cannot communicate directly with one another, but the host can proxy a connection between them.

Here are the main components for the case of Median:

sequenceDiagram
    autonumber
    participant relayer as Relayer (plugin)
    participant core as Plugin (host)
    participant median as Median (plugin)

    Note over core: KeystoreServer
    core->>+relayer: NewRelayer(Config, Keystore)
    Note over relayer: KeystoreClient
    Note over relayer: RelayerServer
    relayer->>-core: Relayer ID 
    Note over core: RelayerClient

    core->>+relayer: NewMedianProvider(RelayArgs, PluginArgs)
    Note over relayer: MedianProviderServer
    relayer->>-core: MedianProvider ID
    Note over core: MedianProvider (Proxy)

    Note over core:  DataSourceServer
    Note over core:  ErrorLogServer

    core->>+median: NewMedianFactory(MedianProvider, DataSource, ErrorLog)
    Note over median: MedianProviderClient
    Note over median: DataSourceClient
    Note over median: ErrorLogClient
    Note over median: MedianFactoryServer
    median->>-core: MedianFactory ID
    Note over core: MedianFactoryClient

    core->>+median: NewReportingPlugin(ReportingPluginConfig)
    Note over median: ReportingPluginServer
    median->>-core: ReportingPlugin ID
    Note over core: ReportingPluginClient

Note: MedianProvider includes multiple component services on the same connection.

sequenceDiagram
    autonumber
    participant relayer as Relayer (plugin)
    participant core as Plugin (host)
    participant median as Median (plugin)

    core->>+relayer: NewMedianProvider(RelayArgs, PluginArgs)
    Note over relayer: OffchainConfigDigesterServer
    Note over relayer: ContractConfigTrackerServer
    Note over relayer: ContractTransmitterServer
    Note over relayer: ReportCodecServer
    Note over relayer: MedianContractServer
    Note over relayer: OnchainConfigCodecServer
    
    relayer->>-core: MedianProvider ID
    Note over core: MedianProvider (Proxy)
    
    Note over core: OffchainConfigDigesterClient
    Note over core: ContractConfigTrackerClient
    Note over core: ContractTransmitterClient
    
    core->>+median: NewMedianFactory(MedianProvider, DataSource, ErrorLog)
    Note over median: ReportCodecClient
    Note over median: MedianContractClient
    Note over median: OnchainConfigCodecClient

Auto-Recovery

The pluginService type contains reusable automatic recovery code.

type pluginService[P grpcPlugin, S services.Service] struct

Each plugin implements their own interface (Relayer, Median, etc.) with a new type that also embeds a pluginService. This new service type implements the original interface, but internally manages re-starting and re-connecting to the plugin as an implementation detail. So there is one long-lived instance of each plugin service and client, and whenever the plugin process (along with its server) crashes, the service restarts the plugin process and re-establishes the client's connection to the new server.

sequenceDiagram
    participant core as Plugin (host)
    participant relayer as Process 1 (plugin)
    participant relayer2 as Process 2 (plugin)
    
    Note over core: RelayerService
    
    core->>+relayer: exec.Cmd
    core->>relayer: NewRelayerServer
    Note over relayer: RelayerServer
    relayer->>core: RelayerID
    Note over core: RelayerClient.init()
    
    loop
        core->>relayer: Alive?
        relayer->>core: Yes
    end
    core->>relayer: Alive?
    relayer->>-core: No

    core->>+relayer2: exec.Cmd
    core->>relayer2: NewRelayerServer
    Note over relayer2: RelayerServer
    relayer2->>core: RelayerID
    Note over core: RelayerClient.update()

    loop
        core->>relayer2: Alive?
        relayer2->>-core: Yes
    end